Time per task: chat reply

A short assistant turn: 400 tokens of context in, 300 tokens out. Time = time to first answer + output tokens ÷ output speed.

As of 25 Oct 2026, BenchLeader ranks Celeris-1 first on the Time per task: chat reply board at 0.7 s, ahead of Gemini 2.5 Flash-Lite (1.3 s) and Gemini 2.5 Flash (1.8 s).

520 models · data as of 25 Oct 2026

1. 1Celeris-10.7 s
2. 2Gemini 2.5 Flash-Lite (none)1.3 s
3. 3Gemini 2.5 Flash (none)1.8 s
4. 4DeepSeek V4.1 Flash (none)1.9 s
5. 5Nemotron 3 Nano 30B A3B (none)1.9 s
6. 6Nova Micro1.9 s
7. 7GPT-4.1 nano2.0 s
8. 8Ministral 3 3B2.1 s
9. 9Gemini 3 Flash (none)2.3 s
10. 10Qwen3 Coder 30B A3B Instruct2.3 s
11. 11Gemini 3.5 Flash (minimal)2.4 s
12. 12GPT-5.4 Mini (none)2.4 s
13. 13GPT-5.4 Nano (none)2.4 s
14. 14Llama 3.2 3B Instruct2.4 s
15. 15Llama 3.2 1B Instruct2.5 s

SearchAll providersAlibabaAmazonAnt GroupAnthropicArcee AIByteDanceCelerisCohereDeepSeekGoogleIBMInceptionMBZUAIMeituanMetaMicrosoftMiniMaxMistral AIMoonshot AIMultiverse ComputingNous ResearchNVIDIAOpenAIReflectionRekaSapiens AISarvamSpaceXAIStepFunSwiss AIThinking MachinesUpstageXiaomiZhipu AIOpen weightsNew modelsBest per model (244)244 of 244

#	Model	seconds ↓	Index ↕
1	DeepSeek V4 Flash 0420maxDeepSeekopen	2.1 min	57.0
2	Qwen3.6 PlusAlibaba	1.8 min	58.8
3	Ling Flash 2.0Ant Groupopen	1.2 min	44.4
4	DeepSeek R1 0528DeepSeekopen ↗	1.2 min	49.5
5	GPT-5.3-CodexxhighOpenAI	1.1 min	63.8
6	Seed-OSS-36B-InstructByteDanceopen ↗	1.1 min	46.8
7	Qwen3.8 2.4T A95BAlibabaopen ↗	56.4 s	61.0
8	MiniMax M2.7MiniMaxopen ↗	52.3 s	55.6
9	GPT-5.4 ProOpenAI	48.9 s	63.3
10	Qwen3.8-Flash-NextAlibabaopen ↗	46.8 s	60.3
11	Kimi K2.7 CodeMoonshot AIopen ↗	45.0 s	55.8
12	MiMo-V2.6-ProXiaomiopen ↗	44.7 s	64.0
13	MiMo-V2.6-FlashXiaomiopen ↗	42.8 s	59.7
14	Ling 3.0 TinyAnt Groupopen	42.7 s	48.7
15	Qwen3.7 PlusAlibaba	41.1 s	58.5
16	North Mini CodeCohereopen	38.6 s	45.9
17	Grok Build 0.1SpaceXAI	36.9 s	54.8
18	MiniMax M2.5MiniMaxopen ↗	36.8 s	53.6
19	GLM 5.3 FlashZhipu AIopen ↗	36.7 s	62.7
20	MiniMax-M2.1MiniMaxopen ↗	35.6 s	53.8
21	Grok 3 minithinkingSpaceXAI	35.3 s	49.4
22	NewKimi K3.1 ProMoonshot AI	34.3 s	66.9
23	LongCat 2.0Meituanopen ↗	32.7 s	49.4
24	GLM-4.5-AirZhipu AIopen ↗	32.5 s	46.6
25	MiMo-V2.5Xiaomiopen ↗	31.9 s	54.3
26	Granite 4.2 30BIBMopen ↗	31.5 s	48.7
27	Magistral Small 1.2Mistral AIopen	30.9 s	40.3
28	GLM 5.3lowZhipu AIopen ↗	28.9 s	55.6
29	Qwen3.5-4Bno reasoningAlibabaopen ↗	28.0 s	42.2
30	Reka Flash 3Rekaopen ↗	27.6 s	38.9
31	Mistral Large 4Mistral AIopen	27.5 s	56.7
32	Granite 4.2 8BIBMopen ↗	27.3 s	46.1
33	MiniMax M3MiniMaxopen ↗	27.2 s	56.4
34	o4-minihighOpenAI	27.2 s	50.1
35	Claude Opus 4.1thinkingAnthropic	27.1 s	56.4
36	MiniMax-M1MiniMaxopen	26.0 s	47.5
37	NewMiniMax M3.1 FlashMiniMaxopen ↗	25.4 s	61.9
38	Gemini 2.5 ProGoogle	24.9 s	53.9
39	MiniMax-M2MiniMaxopen ↗	24.6 s	50.8
40	Granite 4.2 3BIBMopen ↗	23.9 s	44.8
41	Gemini 3.1 ProGoogle	23.3 s	60.5
42	Ring-2.6-1TAnt Groupopen	22.8 s	49.3
43	GPT-5.5 ProxhighOpenAI	22.5 s	65.3
44	DeepSeek V3.1thinkingDeepSeekopen ↗	21.9 s	48.1
45	Agnes 2.5 Pro BetaSapiens AI	21.7 s	59.1
46	Kimi K2 ThinkingthinkingMoonshot AIopen ↗	20.9 s	52.4
47	GLM 5 TurboZhipu AI	20.4 s	55.5
48	Granite 3.3 8Bno reasoningIBMopen ↗	20.2 s	39.2
49	DeepSeek V3.2DeepSeekopen ↗	19.7 s	56.6
50	GLM-4.5Zhipu AIopen ↗	19.7 s	51.7
51	DeepSeek R1DeepSeekopen ↗	19.5 s	46.8
52	Step 3.5 FlashStepFunopen ↗	19.2 s	53.7
53	Step 3.5 Flash 0202StepFunopen	18.8 s	50.7
54	Muse Spark 1.2xhighMeta	18.6 s	63.2
55	Solar Pro 3Upstage	17.9 s	43.2
56	Mistral Medium 3.5Mistral AIopen	16.7 s	49.4
57	Nemotron 3 Super 120B A12BthinkingNVIDIAopen ↗	16.2 s	50.6
58	Muse Spark OpenxhighMetaopen ↗	16.0 s	62.0
59	Command A+Cohereopen	16.0 s	50.5
60	Step 3.7 FlashStepFunopen ↗	15.9 s	51.5

Show 100 more

Cite as: BenchLeader, “Time per task: chat reply”, data as of 25 Oct 2026, https://www.benchleader.com/leaderboards/time-chat-turn.